A predictive maintenance system for medical X-ray equipment

Abstract

This invention discloses a predictive maintenance system for medical X-ray equipment, belonging to the field of medical equipment maintenance technology. It includes a multi-source data acquisition module, a feature construction module, a coupled degradation modeling module, a risk assessment module, and a maintenance strategy generation module. Based on voltage distortion rate, current harmonic components, and temperature field gradient, this invention extracts the standard deviation of voltage distortion rate, the total distortion coefficient of current harmonics, and the rate of change of temperature gradient after denoising using wavelet thresholding, and fuses them into a feature vector. Combining a long short-term memory network and a Bayesian network, it calculates a comprehensive degradation index of multi-parameter coupling; subsequently, it constructs a multi-parameter feature space and outputs a comprehensive risk level through unsupervised learning methods; finally, it generates targeted maintenance strategies based on the risk level and optimization algorithms. This invention achieves accurate prediction and intelligent maintenance of medical X-ray equipment faults, reduces sudden failures, lowers maintenance costs, and improves equipment operational reliability and diagnostic and treatment safety.

Description

Technical Field

[0001] This invention relates to the field of medical equipment maintenance technology, and more specifically, to a predictive maintenance system for medical X-ray equipment. Background Technology

[0002] The maintenance of existing medical X-ray equipment largely relies on a passive approach of periodic inspections or post-failure repairs, which has the following drawbacks: First, data acquisition is limited, relying heavily on simple parameters from the equipment's built-in sensors (such as surface temperature and operating time), making it difficult to reflect the potential degradation status of core components (such as high-voltage generators and cooling systems). Second, it fails to consider the coupling relationships between parameters, such as the electromagnetic induction relationship between voltage distortion and current harmonics, and the thermal conduction relationship between abnormal current and temperature rise, leading to a one-sided degradation assessment. Third, risk assessment depends on human experience and lacks quantitative standards, making it prone to misjudgment or omission. Fourth, maintenance strategies are highly generalized and fail to be dynamically adjusted based on the real-time status of the equipment, potentially leading to over-maintenance or under-maintenance. These problems result in a high probability of sudden equipment failure, high maintenance costs, and even affect the continuity and safety of diagnosis and treatment. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a predictive maintenance system for medical X-ray equipment failures, which addresses the problems of single data, one-sided assessment, subjective risk judgment, and rigid strategies mentioned in the background art through the following solutions.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a predictive maintenance system for medical X-ray equipment, comprising:

[0005] Multi-source data acquisition module: used to acquire multi-source data during the operation of medical X-ray equipment and transmit it to the feature construction module;

[0006] The multi-source data includes voltage distortion rate, current harmonic components, and temperature field gradient.

[0007] Feature construction module: connected to the multi-source data acquisition module, used to perform noise suppression and feature extraction on the multi-source data to generate feature vectors;

[0008] Coupled degradation modeling module: connected to the feature construction module, based on the feature vector, calculates the comprehensive degradation index by establishing a multi-parameter degradation model;

[0009] Risk assessment module: connected to the coupled degradation modeling module, constructs a multi-parameter feature space based on comprehensive degradation index, and uses unsupervised learning method to classify and recognize the multi-parameter features. It classifies and outputs the comprehensive risk level by the distribution of patterns and the degree of abnormal deviation in the feature space.

[0010] Maintenance strategy generation module: Connected to the risk assessment module, it generates corresponding maintenance strategies and decision instructions based on the comprehensive risk level and multi-parameter feature space through optimization algorithms.

[0011] The technical effects and advantages of this invention are as follows:

[0012] This invention uses the Pockels effect, giant magnetoresistive sensing, and infrared-FBG fusion technology to accurately acquire three core parameters: voltage, current, and temperature, covering the electromagnetic and thermal states of equipment. It overcomes the shortcomings of traditional single data and provides a comprehensive basis for degradation assessment.

[0013] This invention quantifies the coupling relationships between parameters, such as electromagnetic induction and thermal conduction, and combines the temporal fitting ability of long short-term memory networks with the probabilistic reasoning ability of Bayesian networks. The output comprehensive degradation index can truly reflect the overall degradation status of the equipment, avoiding the one-sidedness of single parameter evaluation.

[0014] This invention constructs a multi-parameter feature space and classifies it through unsupervised learning. It combines pattern deviation measurement to quantify risk and refines the risk level into 5 levels, solving the problems of subjectivity and vague standards in traditional manual judgment and providing a quantitative basis for maintenance decisions.

[0015] This invention generates targeted maintenance strategies based on risk levels and parameter anomalies, and selects the optimal operation combination through a cost-benefit algorithm to avoid over-maintenance or under-maintenance, reduce the probability of sudden failures, and improve the reliability of equipment operation and the safety of diagnosis and treatment. Attached Figure Description

[0016] Figure 1 This is a system architecture diagram of the present invention;

[0017] Figure 2 This is a system function flowchart of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figure 1 and Figure 2 The illustrated medical X-ray equipment fault prediction maintenance system includes:

[0020] Multi-source data acquisition module: used to acquire multi-source data during the operation of medical X-ray equipment and transmit it to the feature construction module;

[0021] The multi-source data includes voltage distortion rate, current harmonic components, and temperature field gradient:

[0022] Feature construction module: connected to the multi-source data acquisition module, used to perform noise suppression and feature extraction on the multi-source data to generate feature vectors;

[0023] Coupled degradation modeling module: connected to the feature construction module, based on the feature vector, calculates the comprehensive degradation index by establishing a multi-parameter degradation model;

[0024] Risk assessment module: connected to the coupled degradation modeling module, constructs a multi-parameter feature space based on comprehensive degradation index, and uses unsupervised learning method to classify and recognize the multi-parameter features. It classifies and outputs the comprehensive risk level by the distribution of patterns and the degree of abnormal deviation in the feature space.

[0025] Maintenance strategy generation module: Connected to the risk assessment module, it generates corresponding maintenance strategies and decision instructions based on the comprehensive risk level and multi-parameter feature space through optimization algorithms.

[0026] It should be specifically noted that the function of the multi-source acquisition module is to acquire multi-source data during the operation of medical X-ray equipment and transmit it to the feature construction module.

[0027] It should be further explained that the multi-source data includes voltage distortion rate, current harmonic components, and temperature field gradient. The reason for selecting these three sets of data is that voltage distortion rate reflects the transient stability of the output voltage of the high-voltage generator and is an important parameter for assessing whether the equipment has voltage fluctuations or abnormalities; current harmonic components can reflect the degree of distortion of the current waveform and are an important basis for judging whether the equipment has current abnormalities; temperature field gradient can reflect the heat dissipation efficiency of the cooling system and is an important parameter for assessing whether the equipment has overheating risks.

[0028] It should be further explained that the voltage distortion rate is obtained based on the Pockels effect, a physical phenomenon where the refractive index of certain crystals changes linearly with the electric field strength under the influence of an electric field. The distortion degree of the high-voltage generator's output voltage is indirectly obtained through optical measurement. The tools used include an optical fiber probe (containing a Pockels effect crystal, such as lithium niobate (LiNbO3), used to sense the electric field and convert it into an optical signal), a laser source (providing stable incident light), a photodetector (converting the optical signal into an electrical signal), and a signal processor (analyzing and calculating the electrical signal). First, the Pockels crystal in the optical fiber probe is placed in the electric field at the output of the high-voltage generator. The crystal's refractive index *n* changes with the electric field strength *E*, with the following relationship: ( The refractive index is the value when there is no electric field. (This refers to the Pockels coefficient of the crystal, which is related to the crystal material). Then comes optical signal modulation. When the laser passes through the crystal, the change in refractive index causes a phase change in the light, which is transmitted through an optical fiber to a photodetector and converted into a voltage-related electrical signal. The relationship between voltage U and electric field strength E is: d represents the distance between the two electrodes of the crystal; finally, the electrical signal is analyzed by a signal processor to reconstruct the real-time voltage waveform output by the high-voltage generator. Voltage distortion rate is the total harmonic distortion rate of voltage (THD). The effective value of each harmonic voltage is defined as the ratio of the effective value of the fundamental voltage to the effective value of each harmonic voltage. The formula is: ,in This is the effective value of the fundamental voltage (50Hz, the voltage component of the device's operating frequency). The effective values ​​of the 2nd, 3rd...mth harmonic voltages (voltage components whose frequency is an integer multiple of the fundamental frequency).

[0029] It should be further explained that the method for obtaining the current harmonic components utilizes the giant magnetoresistive effect, i.e., the phenomenon that changes in magnetic field cause significant changes in material resistance. The current value is deduced by measuring the magnetic field generated by the current, and then each harmonic component is separated. Specifically, a giant magnetoresistive sensor array, a magnetic field shield, a signal conditioning circuit (converting the resistance change into a measurable voltage signal), and a spectrum analyzer (performing frequency domain analysis of the current signal) are used. First, the magnetic field is measured. The current-carrying conductors of high-voltage equipment generate a surrounding magnetic field. The relationship between the magnetic field strength B and the current I follows Ampere's circuital law: ,in The permeability of free space, The number of turns of the conductor. The distance between the sensor and the wire is given; then comes the resistance-to-current conversion, where the resistance R of the giant magnetoresistive sensor changes with the magnetic field B, i.e. ,in Zero magnetic field resistance The sensitivity coefficient (s) is an inherent characteristic parameter of the giant magnetoresistive sensor, determined through calibration experiments during sensor production: the sensor is placed in a standard magnetic field of known strength, and its resistance change is measured. The relationship between resistance change and magnetic field strength is then established. The s-value is calculated, and the resistance change is converted into a voltage signal through a signal conditioning circuit, from which the real-time current waveform is derived. Finally, harmonic separation is performed, using a spectrum analyzer to analyze the current waveform. A Fourier transform is performed to decompose the fundamental frequency and its harmonic components. The current harmonic components are calculated using Fourier series expansion, resulting in the current waveform. It can be represented as: ,in The current amplitudes are the fundamental frequency, the 2nd harmonic, ..., the mth harmonic. The fundamental angular frequency ( , (fundamental frequency) This represents the phase angle of each harmonic; in practical applications, it is necessary to extract the effective value of each harmonic. , () is used as the characteristic value of the current harmonic component.

[0030] It should be further explained that the temperature field gradient is obtained by fusing infrared thermal imaging technology (large-area temperature distribution measurement) with fiber Bragg grating (FBG) sensing technology (high-precision point temperature measurement). The gradient distribution of the spatial temperature field is obtained through a data fusion algorithm. Specifically, the intensity of infrared radiation emitted from the surface of the device is detected by a high-precision infrared thermal imager, converted into temperature values, and a two-dimensional temperature matrix is ​​output. (x, y are planar coordinates); Fiber Bragg grating measurement center reflected wavelength

[0031] It changes linearly with temperature T, that is... ,in Reference temperature The center wavelength below, The temperature sensitivity coefficient is obtained by demodulating the wavelength shift to obtain the point temperature. (Three-dimensional coordinates); then, using the high-precision point temperature of the fiber Bragg grating as a reference, the measurement error of infrared thermal imaging is corrected, and a three-dimensional temperature field model is constructed through interpolation algorithms. The temperature gradient is the rate of change of temperature in three dimensions of space; it is a vector quantity. The formula is: Temperature Gradient , in the formula , , (The unit is ℃ / m, degrees Celsius per meter, representing the rate of temperature change per unit distance along the x-axis.) These are the partial derivatives of temperature along the x, y, and z axes, respectively. In actual calculations, the ratio of the temperature difference to the distance between adjacent points is used as an approximation, such as... , The gradient is the spatial step size; its magnitude reflects the severity of temperature change, and its direction points in the direction of the fastest temperature increase, which can be used to determine the local overheating trend of equipment.

[0032] It should be specifically explained that the function of the feature construction module is to use the wavelet thresholding method to suppress noise in the voltage distortion rate, current harmonic components, and temperature field gradient. Then, based on the denoised data, the standard deviation of voltage distortion rate, the total distortion coefficient of current harmonics, and the rate of change of temperature gradient are calculated respectively. Finally, through feature fusion, they are integrated into a feature vector.

[0033] It should be further explained that the noise suppression process is as follows: For voltage distortion rate, a db4 wavelet basis suitable for non-stationary signal processing is selected, decomposed into 5 layers, and then a soft thresholding function is applied to the high-frequency coefficients of each layer to filter out the pulse noise introduced by the high-voltage electric field interference. Specifically, for current harmonic components, a sym8 wavelet basis is selected and decomposed into 4 layers, and a hard thresholding function is applied to the high-frequency coefficients to eliminate electromagnetic interference noise in magnetic field measurement. For temperature field gradient, wavelet thresholding is applied along the three spatial dimensions respectively. After decomposition using the coif5 wavelet basis, a hybrid thresholding strategy is adopted to suppress spatial noise during the fusion of infrared thermal imaging and FBG sensing. After processing, the three denoised data sequences are obtained through wavelet reconstruction.

[0034] It should be further explained that the threshold calculation method is based on the high-frequency coefficients of the j-th layer obtained from the decomposition, where the threshold is... ,in The noise standard value, with units consistent with the high-frequency coefficient, is estimated by the median absolute deviation of the highest-level high-frequency coefficient after decomposition. ( (High-frequency coefficients of the highest layer) The number of high-frequency coefficients in the j-th layer.

[0035] It should be further explained that the soft threshold function processing method is applied to the high-frequency coefficients after voltage distortion rate decomposition. The formula is ,in The sign function, by shrinking coefficients exceeding a threshold, can filter out impulse noise introduced by the high-voltage electric field while preserving the subtle fluctuation characteristics of the voltage distortion rate; the hard threshold function processing method is based on the high-frequency coefficients after the decomposition of the current harmonic components. The formula is Directly removing coefficients below a threshold can effectively eliminate sharp noise caused by electromagnetic interference and avoid noise interference with harmonic component extraction. The hybrid threshold strategy processes the high-frequency coefficients after temperature field gradient decomposition along the three-dimensional space (x, y, z). The processing formula is as follows: For strong noise ( ) uses soft threshold shrinkage for weak noise ( This method directly preserves the temperature gradient, which can suppress spatial noise during the fusion of infrared and FBG while retaining the detailed changes in temperature gradient.

[0036] It should be further explained that the high-frequency coefficients are the key components obtained after wavelet decomposition. In wavelet decomposition, the low-frequency coefficients mainly reflect the overall trend of the signal, while the high-frequency coefficients contain the details of signal changes and noise. By analyzing the high-frequency coefficients, noise and signal details can be separated.

[0037] It should be further explained that, based on the denoised data, the standard deviation of the voltage distortion rate is calculated by assuming the denoised voltage distortion rate time series is... (n is the number of sampling points within the time window), standard deviation The formula is: ,in The average value within the time window reflects the degree of fluctuation and dispersion of the voltage distortion rate; the calculation method for the total current harmonic distortion coefficient is to assume that the effective value of the fundamental current is... The jth time ( The effective value of harmonic current is ,in Let be the amplitude of the j-th harmonic current, and then calculate the square root of the sum of the squares of the effective values ​​of each harmonic current, using the formula: Finally, the total current harmonic distortion coefficient is calculated. A higher CHDC value indicates a more severe current distortion; the calculation method for the temperature gradient change rate is to assume... The magnitude of the temperature gradient vector at time step is , Time for Then the rate of change .

[0038] It should be further explained that the feature vector is obtained by taking the feature parameter x (i.e. , , The data is normalized and mapped to the interval [0, 1], resulting in the normalized data. ,in , The minimum and maximum values ​​of this feature parameter in historical data are then used to fuse the three normalized features into a feature vector. ,in , , These are the normalized standard deviation of voltage distortion rate, the total distortion coefficient of current harmonics, and the rate of change of temperature gradient, respectively, with a vector dimension of 1×3.

[0039] It should be specifically explained that the function of the coupled degradation modeling module is to perform correlation analysis on each feature parameter based on feature vectors, quantify the coupling correlation strength between parameters, then preset the basic correlation coefficient of each parameter in degradation assessment based on the physical characteristics and fault mechanism of the equipment, and finally input the multi-parameter features after correlation analysis into the model based on the fusion model of long short-term memory network and Bayesian network, and output a comprehensive degradation index through time-series degradation trend fitting and multi-parameter probability coupling calculation.

[0040] It should be further explained that the method for quantifying the strength of coupling correlation is based on feature vectors. The coupling strength between parameters is quantified using a dual-index approach. Specifically, the linear correlation between any two parameters a and b is first calculated. The formula is: in, The covariance (N is the sample size) is... For the i-th sample value, (sample mean) These are the standard deviations of the two parameters. The larger the absolute value, the stronger the linear coupling; then the degree of nonlinear correlation between parameters is calculated, using the following formula: in, Let be the joint probability distribution of the two parameters. It is a marginal probability distribution (obtained by interval statistics of historical sample data). A larger value indicates a more significant nonlinear coupling; finally, the two correlation levels are weighted and fused to form a 3×3 coupling coefficient matrix. : in, The maximum value of mutual information entropy (used for normalization to) ), diagonal elements (Parameter self-association), non-diagonal elements It reflects the overall coupling strength between the two parameters.

[0041] It should be further explained that the method for presetting the basic correlation coefficients of each parameter in the degradation assessment is based on the electromagnetic coupling characteristics and heat conduction laws of medical X-ray equipment, and the presetting basic correlation coefficient matrix... (3×3), the inherent relationship between quantification parameters: voltage and current ( ): (Voltage distortion and current harmonics in the high-voltage generator are strongly coupled by electromagnetic induction.)

[0042] Voltage and temperature : (The increase in power consumption caused by voltage anomalies has a relatively weak impact on temperature.)

[0043] Current and temperature : (The Joule heating generated by current harmonics has a significant impact on the temperature field.)

[0044] diagonal elements (Self-association);

[0045] By fusing data-driven and mechanism-driven correlation information through matrix dot product, a dynamic correlation coefficient matrix is ​​obtained: (element-wise multiplication, i.e.) ).

[0046] It should be further explained that the comprehensive degradation index is obtained by first sorting the feature vectors according to a time series. (t is the time step) Input the Long Short-Term Memory network, learn the long-term dependencies of the parameters, and output the time-series predicted values ​​of each parameter: , , in, Network parameters (optimized through training with historical time-series data); then using the dynamic correlation coefficient matrix For the conditional probability table, a Bayesian network is constructed, and the time-series predicted values ​​are used as input nodes to calculate the joint degradation probability of multiple parameters: Where F represents a device malfunction event. This represents the probability of device degradation due to parameter coupling at time t (calculated via Bayesian conditional probability propagation); finally, the time-series predicted value and the joint degradation probability are fused to output a comprehensive degradation index. in, The mean of the time-series predicted values ​​(reflecting the time-series degradation trend) is 0.7, and 0.3 are weights calibrated based on historical fault data, determined by minimizing the prediction error; finally... The closer the value is to 1, the more severe the equipment degradation.

[0047] It should be specifically explained that the function of the risk assessment module is to construct a multi-parameter feature space by analyzing the interrelationships and combination patterns among the feature parameters in the comprehensive degradation index; then, to classify and recognize the multi-parameter features using unsupervised learning methods, and output the comprehensive risk level through the distribution of patterns and the degree of abnormal deviation in the feature space.

[0048] It should be further explained that the construction method of the multi-parameter feature space is based on the interrelationship of each feature parameter in the comprehensive degradation index and the comprehensive degradation index. Construct a 4-dimensional feature space: The first three dimensions are the normalized original feature parameters (preserving the independent characteristics of each parameter), and the fourth dimension is the comprehensive degradation index (reflecting the overall degradation state after parameter coupling). This space includes the combination patterns between parameters (such as...). and (The synchronous increase) also integrates the overall degradation level.

[0049] It should be further explained that the K-means algorithm in unsupervised learning is used to classify the samples in the feature space into patterns. The specific steps are as follows: first, initial cluster centers are set; then, based on the historical normal operation data of the equipment (fault-free state), the mean vector of normal samples in the feature space is calculated. Based on the typical characteristics of "slight degradation," "moderate degradation," and "severe degradation" in historical fault data, three abnormal cluster centers are preset (a total of four cluster centers, corresponding to four patterns); then, sample clustering iteration is performed, and pattern division is achieved by minimizing the K-means objective function. in, This represents the k-th cluster (corresponding to a pattern). Let be the center vector of this cluster. Using Euclidean distance, the cluster centers are iteratively updated until convergence, ultimately dividing the samples in the feature space into four patterns: normal pattern. Mild risk mode Medium-risk model High-risk mode .

[0050] It should be further explained that the comprehensive risk level is obtained by first statistically analyzing the current sample. To which cluster And calculate the failure probability of this cluster in historical data. (like Corresponding to a historical failure probability ≥80%, It ranges from 30% to 80%. (e.g., 5%–30%), then calculate the Euclidean distance between the current sample and the center of its respective pattern to quantify the degree of deviation: in, Let be the center vector of the k-th pattern. The larger the value, the more significant the deviation of the sample from the typical characteristics of its pattern, and the higher the potential risk. Finally, combining the pattern category and the degree of abnormal deviation, a five-level comprehensive risk level is determined: Level 1 (No risk): and ( Level 2 (minor risk): (deviation threshold from normal mode) and ,or and Level 3 (Medium Risk): and ,or and Level 4 (High Risk): and ,or and Level 5 (Urgent Risk): and ;in, Deviation threshold set based on historical data ( The 95th percentile of the normal sample deviating from the center. The 80th percentile of the normal sample deviating from the center The 70th percentile of the normal sample deviating from the center. (60th percentile of normal sample deviating from the center).

[0051] It should be specifically noted that the function of the maintenance strategy generation module is to receive the comprehensive risk level output by the risk assessment module, and based on the assessment information reflected in the risk assessment by each characteristic parameter, evaluate and combine the executable maintenance operations through a cost-benefit optimization algorithm, and generate a structured maintenance strategy and decision instructions that match the comprehensive risk level and multi-parameter feature space based on the equipment status and risk changes.

[0052] It should be further explained that the maintenance strategy is generated by first establishing a maintenance operation library and feature relationship mapping, including: based on the core components of medical X-ray equipment (such as high-voltage generators, cooling systems, and sensors), a pre-defined standardized maintenance operation library is established, covering three types of basic operations and combined operations: voltage-related operations; (Insulation testing of high-voltage generator) (Voltage regulation module calibration); Current-related operations: (Giant magnetoresistive sensor array calibration) (Replacing the filter capacitor) Temperature-related operations: (Cleaning the cooling fan) (Heat sink cleaning); Combined operations: such as (Execute simultaneously) and This is used for multi-parameter anomaly scenarios; based on the improvement effect of operations on parameters in historical maintenance data, a correlation coefficient matrix is ​​established. (6×3 matrix), quantization operation pairs (Standard deviation of voltage distortion rate) (Total current harmonic distortion coefficient) Targeted effects of (temperature gradient change rate): ( right (The most effective improvement) (right (weak impact) (right (Minimal impact) ( right (The most effective improvement) (right (There is an indirect impact). ( right The improvement effect is strongest for the parameter with the highest coefficient of 0.1. The higher the coefficient in the matrix, the more significant the abnormal correction effect of the operation on the corresponding parameter.

[0053] It should be further explained that subsequent priority assessments are performed based on risk level and parameter anomalies, including: determining the maintenance urgency and operational intensity according to the comprehensive risk level (levels 1-5): Level 1 (no risk): outputting a "continuous monitoring" command, recording the feature vector every 24 hours. Level 2 (Minor Risk): Prioritize low-invasive procedures (such as...) , Level 3 (Medium Risk): Combine 1-2 targeted actions (such as...) to avoid interrupting normal equipment operation; Execute in case of exception Level 4 (High Risk): Initiating high-intensity operations (such as...) , Simultaneously arrange shutdown and inspection; Level 5 (Emergency Risk): Immediately execute emergency operations (such as...) This triggers the equipment's safety shutdown protection; simultaneously, it incorporates the abnormal deviation of parameters in the feature space. (Refer to the risk assessment module) Dynamically adjust operation priorities: If deviation (If voltage parameters are significantly abnormal), then the priority of voltage-related operations will be increased by 20%; if The highest deviation should be used to increase the priority of temperature-related operations.

[0054] It should be further explained that the final optimization algorithm-based combination of operations and generation strategy includes: using a cost-benefit optimization algorithm to evaluate and combine operations, with the goal of "achieving the maximum reduction in risk level with the lowest maintenance cost," specifically defining the operation cost. (Including time costs and consumable costs), such as (Maintenance Unit) The combined operation cost is the sum of the individual operation costs, and a function to improve the risk level is fitted based on historical data. If executed The average risk level subsequently decreased. Level, then optimize the objective function (Maximize benefit-cost ratio), with the constraint that the number of combined operations ≤ 3 (to avoid over-maintenance) and must cover all abnormal parameters (such as...). (At least one current-related operation must be selected in case of an anomaly). After the algorithm outputs the optimal combination of operations, it generates a strategy containing three elements: Operation steps: such as "1. Execute..." 1. Calibrate the sensor; 2. Perform Clean the air duct; 3. Retest after 2 hours. Time window: For example, Level 5 requires "completion within 1 hour", Level 2 can be "completion within 3 days"; Expected goal: For example, "risk level ≤ Level 2 after execution", ".

[0055] It should be further explained that the decision-making instructions are generated by converting structured strategies into machine-executable instructions, such as: Level 5 risk: "[Emergency Instruction] The equipment shall immediately stop and execute..." "After maintenance, high-voltage insulation testing and harmonic distortion rate testing are required before restarting"; Level 2 Risk: "[Planned Instructions] To be executed within this week" Calibrate the voltage module and retest. And upload it to the system.

[0056] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0057] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A medical X-ray apparatus failure prognostic maintenance system, characterized by, include: Multi-source data acquisition module: used to acquire multi-source data during the operation of medical X-ray equipment and transmit it to the feature construction module; The multi-source data includes voltage distortion rate, current harmonic components, and temperature field gradient. The voltage distortion rate is based on the Paulcile effect and is calculated by measuring the change in the crystal refractive index caused by the output voltage of the high voltage generator using an optical fiber probe containing a lithium niobate crystal, combined with a laser source, photodetector, and signal processor. The current harmonic components are calculated using a giant magnetoresistive sensor array. The current is inferred from the changes in the magnetic field, and the calculation is performed separately using a magnetic field shield, signal conditioning circuit, and spectrum analyzer. The temperature field gradient is calculated by combining a high-precision infrared thermal imager and a fiber Bragg grating sensor through a data fusion algorithm. The infrared thermal imager outputs a two-dimensional temperature matrix, and the fiber Bragg grating obtains the point temperature by changing the central reflection wavelength. After correction, a three-dimensional temperature field model is constructed. Feature construction module: connected to the multi-source data acquisition module, used to perform noise suppression and feature extraction on the multi-source data to generate feature vectors; Coupled degradation modeling module: connected to the feature construction module, based on the feature vector, calculates the comprehensive degradation index by establishing a multi-parameter degradation model; The comprehensive degradation index is obtained by pre-setting a basic correlation coefficient matrix based on the electromagnetic coupling characteristics and heat conduction laws of medical X-ray equipment. A dynamic correlation coefficient matrix is ​​obtained through matrix multiplication. The linear correlation degree between any two parameters (standard deviation of voltage distortion rate, total distortion coefficient of current harmonics, and rate of change of temperature gradient) and the nonlinear correlation degree between the parameters are calculated. The two correlation degrees are weighted and fused to form a 3×3 coupling coefficient matrix. Then, the feature vector is first input into a long short-term memory network according to the time series to learn the long-term dependence of the parameters and output the time series predicted values ​​of each parameter. Then, a Bayesian network is constructed using the dynamic correlation coefficient matrix as a conditional probability table. The time series predicted values ​​are used as input nodes to calculate the joint degradation probability of multiple parameters. Finally, the time series predicted values ​​and the joint degradation probability are fused to output the comprehensive degradation index. Risk assessment module: connected to the coupled degradation modeling module, constructs a multi-parameter feature space based on comprehensive degradation index, and uses unsupervised learning method to classify and recognize the multi-parameter features. It classifies and outputs the comprehensive risk level by the distribution of patterns and the degree of abnormal deviation in the feature space. Maintenance strategy generation module: connected to the risk assessment module, it generates corresponding maintenance strategies and decision instructions based on the comprehensive risk level and multi-parameter feature space through optimization algorithms; The maintenance strategy and decision instruction acquisition method is to establish a maintenance operation library, including voltage related operation, current related operation, temperature related operation and combined operation, to determine the maintenance urgency according to the risk level: level 1 outputs "continuous monitoring", level 2 prefers low-invasive operation, level 3 combines 1-2 targeted operations, level 4 starts specified operation and arranges shutdown inspection, level 5 immediately performs emergency operation and triggers safety shutdown; combined with the abnormal deviation degree, the operation priority is dynamically adjusted, the cost-benefit optimization algorithm is used to output the optimal operation combination, and the structured strategy including operation steps, time window and expected target and machine executable instruction is generated.

2. The medical X-ray apparatus failure prognostic maintenance system of claim 1, wherein: The function of the feature construction module is to use wavelet threshold method to suppress noise of voltage distortion rate, current harmonic component and temperature field gradient, then calculate voltage distortion rate standard deviation, current harmonic total distortion coefficient and temperature gradient change rate based on the denoised data, and finally integrate them into a feature vector through feature fusion.

3. The medical X-ray apparatus failure prognostic maintenance system of claim 2, wherein: The wavelet threshold method adapts to different data sources and adopts different processing strategies. For voltage distortion rate, db4 wavelet basis is used to decompose to 5 layers and apply soft threshold function; for current harmonic component, sym8 wavelet basis is used to decompose to 4 layers and apply hard threshold function; for temperature field gradient, coif5 wavelet basis is used to decompose and apply mixed threshold strategy.

4. The medical X-ray apparatus failure prognostic maintenance system of claim 1, wherein: The function of the risk judgment module is to analyze the correlation and combination mode between each feature parameter in the comprehensive degradation index, and construct a multi-parameter feature space; then, use unsupervised learning method to classify and recognize the multi-parameter features, and output the comprehensive risk level through the distribution and abnormal deviation degree of the mode in the feature space.

5. The medical X-ray apparatus failure prognostic maintenance system of claim 4, wherein: The unsupervised learning method is K-means clustering algorithm, that is, the K-means clustering algorithm is used to classify the multi-parameter feature space samples. In the clustering initialization stage, based on the distribution characteristics of the data, the selection of the initial clustering center is optimized. In the clustering process, the Euclidean distance is used as the similarity measure standard between samples, and the distance between samples and clustering center is iteratively calculated until convergence.

Images